Re: [PATCH v2 net-next 04/12] net: ethtool: netlink: retrieve stats from multiple sources (eMAC, pMAC)

[Vladimir Oltean]

On Sun, Jan 15, 2023 at 01:22:14AM +0200, Vladimir Oltean wrote:
> On Fri, Jan 13, 2023 at 08:43:36PM -0800, Jakub Kicinski wrote:
> > On Wed, 11 Jan 2023 18:16:58 +0200 Vladimir Oltean wrote:
> > > +/**
> > > + * enum ethtool_stats_src - source of ethtool statistics
> > > + * @ETHTOOL_STATS_SRC_AGGREGATE:
> > > + *	if device supports a MAC merge layer, this retrieves the aggregate
> > > + *	statistics of the eMAC and pMAC. Otherwise, it retrieves just the
> > > + *	statistics of the single (express) MAC.
> > > + * @ETHTOOL_STATS_SRC_EMAC:
> > > + *	if device supports a MM layer, this retrieves the eMAC statistics.
> > > + *	Otherwise, it retrieves the statistics of the single (express) MAC.
> > > + * @ETHTOOL_STATS_SRC_PMAC:
> > > + *	if device supports a MM layer, this retrieves the pMAC statistics.
> > > + */
> > > +enum ethtool_stats_src {
> > > +	ETHTOOL_STATS_SRC_AGGREGATE,
> > > +	ETHTOOL_STATS_SRC_EMAC,
> > > +	ETHTOOL_STATS_SRC_PMAC,
> > > +};
> > 
> > Should we somehow call it "MAC stats"?
> > 
> > Right now its named like a generic attribute, but it's not part of 
> > the header nest (ETHTOOL_A_HEADER_*).
> > 
> > I'm not sure myself which way is better, but feels like either it
> > should be generic, in the header nest, and parsed by the common code;
> > or named more specifically and stay in the per-cmd attrs.
> 
> Considering that I currently have separate netlink attributes for
> ETHTOOL_MSG_STATS_GET (ETHTOOL_A_STATS_SRC) and for
> ETHTOOL_MSG_PAUSE_GET (ETHTOOL_A_PAUSE_STATS_SRC), I'm going to add just
> a single attribute right under ETHTOOL_A_HEADER_FLAGS for v3 and go from
> there. Is it ok if I keep naming it ETHTOOL_A_STATS_SRC, or would you
> prefer something else?

I'm already lost while trying to implement this change request.

ETHTOOL_A_STATS_HEADER uses NLA_POLICY_NESTED(ethnl_header_policy),
while ETHTOOL_A_PAUSE_HEADER uses NLA_POLICY_NESTED(ethnl_header_policy_stats).

The two header nest policies look like this:

const struct nla_policy ethnl_header_policy[] = {
	[ETHTOOL_A_HEADER_DEV_INDEX]	= { .type = NLA_U32 },
	[ETHTOOL_A_HEADER_DEV_NAME]	= { .type = NLA_NUL_STRING,
					    .len = ALTIFNAMSIZ - 1 },
	[ETHTOOL_A_HEADER_FLAGS]	= NLA_POLICY_MASK(NLA_U32,
							  ETHTOOL_FLAGS_BASIC),
};

const struct nla_policy ethnl_header_policy_stats[] = {
	[ETHTOOL_A_HEADER_DEV_INDEX]	= { .type = NLA_U32 },
	[ETHTOOL_A_HEADER_DEV_NAME]	= { .type = NLA_NUL_STRING,
					    .len = ALTIFNAMSIZ - 1 },
	[ETHTOOL_A_HEADER_FLAGS]	= NLA_POLICY_MASK(NLA_U32,
							  ETHTOOL_FLAGS_STATS),
};

The request seems to be for ETHTOOL_A_PAUSE_HEADER to use a policy like this:

 const struct nla_policy ethnl_header_policy_mac_stats[] = {
 	[ETHTOOL_A_HEADER_DEV_INDEX]	= { .type = NLA_U32 },
 	[ETHTOOL_A_HEADER_DEV_NAME]	= { .type = NLA_NUL_STRING,
 					    .len = ALTIFNAMSIZ - 1 },
 	[ETHTOOL_A_HEADER_FLAGS]	= NLA_POLICY_MASK(NLA_U32,
 							  ETHTOOL_FLAGS_STATS),
+	[ETHTOOL_A_HEADER_MAC_STATS_SRC] = NLA_POLICY_MASK(NLA_U32,
+							   ETHTOOL_MAC_STATS_SRC_PMAC),
 };

and for ETHTOOL_A_STATS_HEADER to use a policy like this:

const struct nla_policy ethnl_header_policy_mac_stats_src_basic[] = {
	[ETHTOOL_A_HEADER_DEV_INDEX]	= { .type = NLA_U32 },
	[ETHTOOL_A_HEADER_DEV_NAME]	= { .type = NLA_NUL_STRING,
					    .len = ALTIFNAMSIZ - 1 },
	[ETHTOOL_A_HEADER_FLAGS]	= NLA_POLICY_MASK(NLA_U32,
							  ETHTOOL_FLAGS_BASIC),
+	[ETHTOOL_A_HEADER_MAC_STATS_SRC] = NLA_POLICY_MASK(NLA_U32,
+							   ETHTOOL_MAC_STATS_SRC_PMAC),
};

Did I get this right?